Abstract: Given the inaccuracy of the regression model of desulphurization efficiency in the operation optimization of desulphurization system, and the difficulty of effective operation of operation optimization scheme in actual operation, his paper presented an optimization method for desulfurization system operation based on cyber-physical fusion and XGBoost-multigroup parallel genetic algorithm. By constructing the XGBoost desulfurization efficiency regression model as the variable of desulfurization variable cost function, MPGA was used to find the corresponding desulfurization efficiency when the variable cost of desulfurization was the lowest, and then the optimal values of the liquid-gas ratio, pH value of absorber slurry and absorber liquid level in the regression model of desulfurization efficiency were obtained. The example analysis show that XGBoost-MPGA had better prediction performance than BP neural network, Random Forest, and GBRT regression model, and had better stability and convergence in the optimization process of desulfurization variable cost extreme value compared with XGBoost-SGA. In addition, the cyber-physical fusion method eliminates the influence of desulfurization physical equipment on the variable cost information of desulfurization after the operation optimization operation and improves the reliability and economy of the operation optimization operation plan.